# Furkan Kupcu

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Furkan Kupcu](https://cerebralvalley.ai/u/furkankupcu)
- **GitHub:** https://github.com/furkankupcu/playwhy
- **Demo video:** https://youtu.be/p1A3SR2Veiw
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1195

PlayWhy turns one video of a football move into an explanation of why it worked, with every claim tied to evidence you can check in the clip.

Professional clubs explain goals with multi-camera tracking data. Amateur coaches, analysts and fans have a single video. PlayWhy is for them.

You upload a clip of up to 20 seconds. The app tracks every player, both keepers and the ball, and maps them from pixels to pitch metres. From those positions it measures the passes, the offside line, who lost his marker, and the numbers that make a goal possible: attackers against defenders around the ball and in the box, free attackers in the final third, and the widest gap in the back line compared with its usual width. It finds the turning point, the moment the attack's edge rose most, and checks whether the nearest defender could still have stopped the free attacker. It also reads what positions cannot show: where key players were facing and how quickly they reacted.

The result opens with a one-line headline giving the main reason, followed by short claims. Each claim cites its evidence, coloured green for an edge for the attack and red for a weakness. The turning point is frozen on the video and the pitch, with the free run and the speed the defender needed. Code drops any claim that cites no evidence, is far from it in time, or quotes a number its evidence does not contain. A second pass strikes unsupported wording, and the page shows what it struck. Viewers can ask their own questions; an answer must cite evidence, or the app says the data cannot tell. Clips without a goal are explained as a move.

We measure ourselves. On a phone video from the stands, calibration is within 1.45 m (median) of 30 hand-marked points. The README lists where it fails, too.

## More from Meta Global AI Developer Hackathon

- [Adrian Gajewski](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1192)
- [Gatekeeper](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1193)
- [Clone Flow](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1194)
- [Max Wo](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1196)
- [Listing Partner](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1197)
- [RegulaGraph](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1198)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1195. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt